Sulfur-modified iron-based biochar via feature-sparsified cascade transfer learning: from interfacial structure...

Area of Science:

• Materials Science
• Environmental Engineering
• Machine Learning

Background:

• Iron-based biochar is effective for phosphorus recovery but suffers from iron aggregation at high loadings.
• Optimizing synthesis parameters for sulfur-doped biochar is challenging due to complex interactions and limited data.

Purpose of the Study:

• To develop a feature-sparsified cascaded transfer learning (FS-CTL) framework for optimizing sulfur-doped iron-based biochar.
• To determine optimal precursor ratios for enhanced phosphorus recovery.

Main Methods:

• Constructed an FS-CTL framework using Cr (VI) removal data as the source domain.
• Applied L1 regularization for feature weight attenuation during transfer fine-tuning.
• Determined optimal S/C and Fe/C precursor ratios.

Main Results:

• Achieved a maximum adsorption capacity of 78.6 mg·g⁻¹ for phosphorus recovery.
• Demonstrated superior performance compared to sulfur-free and high-sulfur controls.
• Confirmed stable removal efficiency across a broad pH range (2.0-9.0).

Conclusions:

• Appropriate sulfur doping inhibits iron aggregation and crystalline growth.
• Sulfur enhances iron binding to the carbon substrate, restricting migration.
• FS-CTL is applicable for cross-task materials design in environmental applications.